ChuanqiXu added a comment. In D129748#3666511 <https://reviews.llvm.org/D129748#3666511>, @tahonermann wrote:
> I can imagine several ways in which things may be going wrong. Perhaps > (de)serialization is switching to a canonical declaration when it should not > be (e.g., reading/writing a definition when only a declaration should be). > Perhaps, for implicitly instantiated declarations (at least for ones for > which `__preferred_name__` is attached), separate non-defining and defining > declarations should be (de)serialized (in the dumps above, there appears to > only be a single address associated with the `basic_string_view<char>` > specialization). Perhaps `ClassTemplateSpecializationDecl` should be > (de)serialized such that a (valid) non-defining declaration is created before > attempting to load anything else that might lead to recursion to the same > declaration (the intent being that an early return when recursion is detected > still yields a valid incomplete declaration). I feel like the problem may come in the process of instantiation. For example, if we add the attribute for `CXXRecordDecl` at `0x7913a18` in this specific case, it would be fine. Although it **may** be OK to address the problem in serialization/deserialization side too, I just feel like it might not be worhy now.. CHANGES SINCE LAST ACTION https://reviews.llvm.org/D129748/new/ https://reviews.llvm.org/D129748 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
